Current B200 cloud pricing
Today’s sourced per-provider B200 on-demand rates, source-linked and dated.
← Back to Compute College
Compute College
B200 Price Per Hour Explained
B200 price per hour is the hourly cost of accessing NVIDIA B200-generation AI compute capacity.
B200 price per hour definition
B200 price per hour is the hourly cost to rent or operate NVIDIA B200-generation AI compute capacity. Each B200 pairs Blackwell-generation compute with large, fast memory — roughly 192 GB of HBM3e at about 8 TB/s, versus 141 GB of HBM3e on the H200 — which can help large-model training and memory-heavy serving. Because newer accelerator capacity can be scarce early in a deployment cycle, the quoted rate may reflect both expected workload value and a premium for access.
Memory trick: B200 pricing is not only a GPU rate; it is a read on how scarce the next compute cycle is.
Live price band
B200 on-demand price band
B200 on-demand capacity ranges roughly $5.89–$6.99 per GPU-hour across 2 sourced providers, as of Jul 7, 2026. Each row below links to the provider's public price page and carries its own observation date.
Public on-demand list prices normalized to a per-GPU-hour rate — not negotiated quotes or reserved pricing. How we label and date evidence: methodology.
Sourced providers
Per-provider B200 on-demand rates
| Provider | Region | $/GPU-hour | Source | Observed |
|---|---|---|---|---|
| RunPod Secure | Multi-region | $5.89 | price page | Jul 7, 2026 |
| Lambda | Multi-region | $6.99 | price page | Jul 7, 2026 |
Why it matters
Why B200 hourly pricing matters
B200 pricing helps ComputeTape readers watch the transition from H100 and H200 capacity into a new infrastructure generation. New systems may change model-training economics, model-serving economics, cluster design, and the competitive position of providers that can deliver usable capacity first.
- Early availability can be valuable to buyers racing to train or serve more demanding models.
- High rates may compensate for scarce systems, infrastructure upgrades, or priority access.
- The premium is a useful signal only when compared with actual workload output and contract terms.
Simple example
Take a hypothetical comparison: H100 capacity rents for $8 per GPU-hour and B200 capacity rents for $16 per GPU-hour. B200 is twice the headline rate. If it completes the same defined workload 2.5 times faster under comparable conditions, the effective cost of that workload could be lower despite the higher hourly price.
- Do not publish an assumed speedup as a fact; measure or source it for a defined job.
- Include networking, system configuration, availability, and runtime in the comparison.
- A scarce accelerator with no available cluster cannot solve a buyer deadline at any advertised rate.
Example figures are illustrative calculations, not current quoted market prices.
Market signal
How to read the market signal
A very high B200 premium can signal early scarcity, strong buyer demand, limited connected-cluster availability, or provider pricing power. If that premium falls over time, it may reflect increased supply, competitive offers, improving deployment, or demand shifting elsewhere.
- Watch deliverable cluster capacity, not only chip announcements.
- Compare premium movement with power, cooling, and data-center readiness.
- Read next-generation rates against H100 and H200 benchmarks to see whether substitution is occurring.
Market read: early B200 pricing can reflect optionality and urgency as much as ordinary rental cost. Follow actual deliverable capacity and comparable workload results before assuming a lasting premium. Figures here are illustrative unless explicitly sourced and dated — see our methodology.
Common mistake
Do not treat a new-generation GPU as automatically cheaper, faster for every job, or economically superior. A valid buyer comparison depends on workload, configuration, memory use, network performance, delivered availability, and how much time or risk the buyer saves.
Practical takeaway
What you can do with this
Treat B200 pricing as both a cost question and an access question. Ask providers what capacity can be delivered, in what configuration, under what term, and with what evidence of performance for the intended training or serving workload.
- Procurement teams: compare deliverable capacity and completed-job economics.
- Analysts and investors: track premiums as an indicator of next-cycle supply tightness, not a standalone valuation claim.
- Infrastructure teams: account for power and cooling readiness before treating systems as usable supply.
- Buyers: distinguish early-access value from a long-lived price assumption when planning future budgets.
- Operators: confirm that software, network fabric, and rack configuration support the advertised system benefit.
Decision check: separate the price of early access from the expected steady-state rate, and require deliverable system capacity before budgeting around it.
Compute College
Turn the lesson into a number
Use the GPU-Hour Cost Calculator, AI Training Cost Calculator, or Model Serving Cost Calculator.
Compute College track
GPU Pricing & Capacity
Step 3 of 8: B200 price per hour